WebAssembly

WebAssembly (often shortened to Wasm) is a low-level bytecode standard supported by all major browsers. It provides a compilation target for languages that compile to machine code — Rust, C, C++, Go, and others — so that performance-sensitive code can run inside a web page without the overhead of JavaScript interpretation. Wasm modules load quickly, execute deterministically, and operate in a sandboxed environment with no direct access to the host system beyond what the browser explicitly exposes.

For AI in the browser, WebAssembly is the infrastructure that makes meaningful client-side inference feasible. Neural network runtimes such as ONNX Runtime Web and TensorFlow.js compile critical paths to WebAssembly, and in many cases offload matrix operations to WebGPU or WebGL for GPU acceleration. The result is that reasonably sized models — stylometric classifiers, sentence-embedding models, even smaller generative models — can run inside a browser tab with acceptable latency and without any server round trips.

The practical implication for privacy-sensitive applications is significant. Tools that analyze potentially sensitive text — legal documents, draft academic papers, private correspondence — can now do their work entirely on the user's device. Coda One's AI Detector does not use WebAssembly. Its text is sent to a server endpoint for scoring; what is not sent is anything beyond the text itself, and what is stored is a word count rather than the words. The scoring is arithmetic over word and sentence statistics, so the cost of running it is close to zero either way.